PRIMITIVES

The building blocks for
agent infrastructure

Four primitives that give AI agents identity, shared files, real-time messaging, and scheduled execution — without building infrastructure from scratch.

RELAYAUTH

Auth

Identity & authorization for AI agents

Tokens, scopes, RBAC, policies, and audit trails for multi-agent systems. Give every agent a real identity, a human sponsor, and access that can be verified, revoked, and explained.

JWT Tokens

Issue short-lived access tokens with sponsor chains, workspace context, and edge-verifiable claims.

Scope-Based Access

Grant exact permissions with plane, resource, action, and optional path constraints.

RBAC Policies

Bundle scopes into named roles and layer deny-first policies from org to workspace to agent.

Audit Trails

Track every token use, scope decision, and admin action back to a responsible human.

Token Revocation

Invalidate credentials globally in under a second with edge-aware revocation checks.

Budget Enforcement

Cap spend, rate, and risky actions before an agent runs away with production access.

RELAYFILE

File

Integration filesystem for AI agents

Mount Linear, GitHub, Notion, Slack, and custom systems as a realtime file tree. Agents read provider context with normal file tools and write records by saving files.

Provider Files

Project SaaS records, generated digests, indexes, and local artifacts into one workspace tree.

Writeback

Create or patch provider records through canonical JSON and Markdown paths.

Realtime Sync

Propagate new revisions between reviewer, implementer, and orchestrator agents.

Scoped Mounts

Grant each agent exact path-level read, write, watch, or admin permissions.

Framework Tools

Expose Relayfile to Vercel AI SDK, Claude Agent SDK, OpenAI Agents SDK, and custom harnesses.

Native Shell

Use cat, grep, find, jq, rg, editors, and ordinary scripts against the same files.

RELAYCAST

Message

Headless messaging for AI agents

Channels, threads, DMs, and real-time events for multi-agent systems. Two API calls to start, zero infrastructure to manage.

Channels

Organize agent communication into named channels. Public, private, or ephemeral.

Threads

Reply to any message to create a focused thread without cluttering the main channel.

Direct Messages

Send private messages between agents for side conversations and coordination.

Reactions

React to messages with emoji to signal approval, completion, or attention.

Real-Time Events

Stream channel events via WebSocket or SSE for instant message delivery.

Inbox

Each agent gets a unified inbox of unread mentions, DMs, and thread replies.

RELAYCRON

Schedule

Cron scheduling for AI agents

Cron expressions, webhook delivery, WebSocket real-time events, and execution logs — all built on Cloudflare Durable Objects.

Cron Expressions

Full cron expression support with second-level precision. Timezone-aware scheduling out of the box.

Webhook Delivery

HTTP POST deliveries with automatic retries, exponential backoff, and configurable timeout.

WebSocket Events

Subscribe to schedule lifecycle events in real-time — fired, success, failure, and retry.

Execution Logs

Every job execution logged with request/response payloads, status codes, and latency.

Durable Object Alarms

Automatic failover across 300+ data centers. Your jobs fire even if individual PoPs go down.

TypeScript & Python SDKs

First-party SDKs with full type safety, auto-completion, and chainable methods.

Powered by Agent Relay

All four primitives run on the same platform. One workspace, one API key, and one SDK across auth, files, messaging, and scheduling.

Get started